Vehicle motion-based driver anger state early warning method and system

Abstract

The invention discloses a driver anger state early warning method and system based on vehicle movement, and the method comprises the following steps: obtaining the historical driving factors of anger mood migration, the historical driving factors including the headstock distance and the trajectory deviation; constructing a Gaussian mixture hidden Markov model based on a Gaussian mixture model and a continuous hidden Markov model, and training the Gaussian mixture hidden Markov model based on the vehicle head distance and the trajectory deviation to obtain an angry state recognition model; and monitoring the driver by using the anger state recognition model, and when the anger level is monitored to be higher than a preset threshold value, intervening the driving behavior of the driver. The method not only can improve the rationality and applicability of the angry state recognition model, but also provides an adaptive decision basis based on the driver emotional state for the driver model used by the traffic vehicle in automatic driving simulation, and further optimizes the man-machine interaction experience of the automatic driving vehicle.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent vehicle human-computer interaction, and in particular relates to a method and system for warning of a driver's anger state based on vehicle motion. Background Art

[0002] This technology is used to estimate the driver's possible "driving rage (road rage)" state through vehicle motion during driving. It can be used to prompt the driver to adjust the state or warn the vehicle's auxiliary driving system to avoid possible dangerous situations.

[0003] The technical defects in the existing technology include: traditional facial expression recognition methods (such as CN202010789012.3) may fail in scenarios with unfavorable lighting conditions or occlusion; anger detection based on a single voice feature (such as volume decibel value) is easily interfered by environmental noise; the static threshold method used in the existing driving emotion patent (US2021123456A1) cannot capture the stage evolution characteristics of the anger state; behavioral analysis based on a fixed time window (such as statistics on the number of sudden accelerations) has a single feature and the results have weak generalization ability. Summary of the Invention

[0004] The present invention aims to address the deficiencies of the prior art and provides the following solution: a method for warning of driver anger based on vehicle motion, comprising the following steps:

[0005] Obtaining historical driving factors of anger emotion transition, wherein the historical driving factors include: headway and trajectory deviation;

[0006] A Gaussian mixture hidden Markov model is constructed based on a Gaussian mixture model and a continuous hidden Markov model, and the Gaussian mixture hidden Markov model is trained based on the headway and trajectory deviation to obtain an anger state recognition model;

[0007] The anger state recognition model is used to monitor the driver, and when the anger level is detected to be higher than a preset threshold, the driver's driving behavior is intervened.

[0008] Preferably, the method for obtaining the headway includes:

[0009] For each vehicle, in each time step, extract the headway data of the vehicle closest to the current vehicle;

[0010] The average of the headway data is calculated within a 2-second time window to obtain the headway distance:

[0011]

[0012] Among them, Avg_Space_Headway 2srepresents the headway distance, N=20 represents the number of frames in a 2-second time window, t0 represents the time of the previous moment, and Space_Headway(t) represents the headway time of the current frame.

[0013] Preferably, the method for obtaining the trajectory deviation includes:

[0014] Select the current vehicle's position coordinates 2 seconds ago (X prev ,Y prev ) and driving speed v prev ;

[0015] Based on the position coordinates (X prev ,Y prev ), calculate the current vehicle's driving direction θ 2 seconds ago:

[0016]

[0017] Based on the position coordinates (X prev ,Y prev ), the driving speed v prev and the driving direction θ, calculate the expected position (X pred ,Y pred ):

[0018] X pred =X prev +v prev ·2·cos(θ)

[0019] Y pred =Y prev +v prev ·2·sin(θ);

[0020] Get the actual position of the vehicle and based on the expected position (X pred ,Y pred ) Calculate the trajectory deviation:

[0021]

[0022] Among them, Deviation represents the trajectory deviation, X t Indicates the horizontal coordinate of the actual position of the vehicle, Y t The vertical coordinate representing the actual position of the vehicle.

[0023] Preferably, the Gaussian mixture hidden Markov model includes: a hidden state and an observation state;

[0024] The hidden state is used to represent the driver's anger level, including 4 discrete states:

[0025] S t∈{S1,S2,S3,S4}

[0026] Among them, S t represents the anger state at time t;

[0027] The observation state is composed of the headway and the trajectory deviation:

[0028] o t =[Avg_Space_Headway(t),Deviation(t)]

[0029] Among them, t represents the observed state of the vehicle at time t.

[0030] Preferably, the training method includes:

[0031] The Gaussian mixture hidden Markov model is trained using the Baum-Welch algorithm;

[0032] Optimize model parameters by maximizing the log-likelihood of the observed data:

[0033]

[0034] Among them, λ represents the model parameter, Indicates the maximum value returned, O represents the sequence of vehicle observation states, and P(O|λ) represents the probability of each observation state occurring under the model parameter λ;

[0035] The model parameters include: initial state probability π i , state transfer matrix A ij And the Gaussian mixture model parameters:

[0036] π i =P(S1=i)

[0037] A ij =P(s t+1 =j|S t =i)

[0038]

[0039] Among them, π i A represents the probability that the driver is in state i at time t=1. ij represents the probability of transitioning from state i to state j, P(o t |S t =j) represents the Gaussian mixture model parameters in state S t =j, Represents the probability density function of multidimensional Gaussian distribution, M represents the number of Gaussian distributions, m represents the mth Gaussian distribution, wj,m represents the weight of Gaussian distribution m in state j, μ j,m represents the mean of the mth Gaussian distribution, Σ j,m represents the covariance of the mth Gaussian distribution, d represents the dimension, T represents the matrix transpose, μ represents the mean, and Σ represents the covariance matrix.

[0040] The present invention also provides a vehicle motion-based driver anger state warning system, the system applying any of the above-mentioned methods, comprising: a historical data acquisition module, a model building module, and a monitoring module;

[0041] The historical data acquisition module is used to obtain historical driving factors of anger emotion transition, and the historical driving factors include: headway and trajectory deviation;

[0042] The model building module builds a Gaussian mixture hidden Markov model based on the Gaussian mixture model and the continuous hidden Markov model, and trains the Gaussian mixture hidden Markov model based on the headway and trajectory deviation to obtain an anger state recognition model;

[0043] The monitoring model uses the anger state recognition model to monitor the driver, and when the anger level is detected to be higher than a preset threshold, the driver's driving behavior is intervened.

[0044] Preferably, in the historical data acquisition module, the process of acquiring the headway distance includes:

[0045] For each vehicle, in each time step, extract the headway data of the vehicle closest to the current vehicle;

[0046] The average of the headway data is calculated within a 2-second time window to obtain the headway distance:

[0047]

[0048] Among them, Avg_Space_Headway 2s represents the headway distance, N=20 represents the number of frames in a 2-second time window, t0 represents the time of the previous moment, and Space_Headway(t) represents the headway time of the current frame.

[0049] Preferably, in the historical data acquisition module, the process of acquiring the trajectory deviation includes:

[0050] Select the current vehicle's position coordinates 2 seconds ago (X prev ,Y prev ) and driving speed v prev ;

[0051] Based on the position coordinates (Xprev ,Y prev ), calculate the current vehicle's driving direction θ 2 seconds ago:

[0052]

[0053] Based on the position coordinates (X prev ,Y prev ), the driving speed v prev and the driving direction θ, calculate the expected position (X pred ,Y pred ):

[0054] X pred =X prev +v prev ·2·cos(θ)

[0055] Y pred =Y prev +v prev ·2·sin(θ);

[0056] Get the actual position of the vehicle and based on the expected position (X pred ,Y pred ) Calculate the trajectory deviation:

[0057]

[0058] Among them, Deviation represents the trajectory deviation, X t Indicates the horizontal coordinate of the actual position of the vehicle, Y t The vertical coordinate representing the actual position of the vehicle.

[0059] Preferably, in the model building module, the Gaussian mixture hidden Markov model includes: a hidden state and an observation state;

[0060] The hidden state is used to represent the driver's anger level, including 4 discrete states:

[0061] S t ∈{S1,S2,S3,S4}

[0062] Among them, S t represents the anger state at time t;

[0063] The observation state is composed of the headway and the trajectory deviation:

[0064] o t =[Avg_Space_Headway(t),Deviation(t)]

[0065] Among them, trepresents the observed state of the vehicle at time t.

[0066] Preferably, in the model building module, the training process includes:

[0067] The Gaussian mixture hidden Markov model is trained using the Baum-Welch algorithm;

[0068] Optimize model parameters by maximizing the log-likelihood of the observed data:

[0069]

[0070] Among them, λ represents the model parameter, Indicates the maximum value returned, O represents the sequence of vehicle observation states, and P(O|λ) represents the probability of each observation state occurring under the model parameter λ;

[0071] The model parameters include: initial state probability π i , state transfer matrix A ij And the Gaussian mixture model parameters:

[0072] π i =P(S1=i)

[0073] A ij =P(s t+1 =j|S t =i)

[0074]

[0075] Among them, π i A represents the probability that the driver is in state i at time t=1. ij represents the probability of transitioning from state i to state j, P(o t |S t =j) represents the Gaussian mixture model parameters in state S t =j, Represents the probability density function of the multidimensional Gaussian distribution, M represents the number of Gaussian distributions, m represents the mth Gaussian distribution, w j,m represents the weight of Gaussian distribution m in state j, μ j,m represents the mean of the mth Gaussian distribution, Σ j,m represents the covariance of the mth Gaussian distribution, d represents the dimension, T represents the matrix transpose, μ represents the mean, and Σ represents the covariance matrix.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] This invention can more intuitively reflect whether a driver's short-term control behavior meets their psychological expectations and, to a certain extent, predict anger triggered by operational constraints or unsatisfactory road conditions. By combining two core observation variables—headway distance, which measures the impact of external pressure, and trajectory deviation, which measures the failure of internal expectations—the two can more comprehensively capture the dynamic evolution of anger. This not only improves the rationality and applicability of anger state recognition models, but also provides a basis for adaptive decision-making based on the driver's emotional state for driver models used in autonomous driving simulations, further optimizing the human-machine interaction experience in autonomous vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0079] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0081] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0082] As a common emotional state, the formation and change of anger are affected by many factors.

[0083] Subjective cognitive evaluation theory, the most widely used theoretical framework, posits that the generation of emotions depends on an individual's cognitive evaluation of the external environment. In the driving environment, anger is often influenced by external factors such as road conditions, driving goals, and traffic flow. This is also influenced by the driver's individual characteristics (such as driving style and personality), which shape specific driving behavior patterns. Therefore, understanding the generation mechanism of anger will help further develop models of driver anger states.

[0084] The evolution of anger in a driving environment can be attributed to the following key processes: Triggering: When a driver encounters an unexpected event (such as the vehicle ahead suddenly braking or cutting in), their emotional state may change. Accumulation: Continuous adverse driving conditions (such as prolonged traffic congestion) may lead to a gradual increase in anger. Dissipation: When the driving environment improves or the driver's psychological state adjusts, anger may gradually subside.

[0085] Therefore, constructing a model of driver anger's emotional transition requires fully considering the triggering mechanism, accumulation process, and relief mechanism of emotion, and incorporating dynamic modeling methods to quantitatively describe the transition of anger. Based on existing research, the driver's anger transition process has the following characteristics: emotional states are dynamic, influenced by the previous emotional state and the external environment; emotional intensity can be quantified using computational models, such as the OCC model or HMM model, which can be used to characterize the transition probability of emotional states; external stimuli (such as traffic congestion and unfulfilled driving goals) are the main factors affecting emotional transition; and individual characteristics (such as driving style and personality) influence the rate and degree of emotional change.

[0086] During driving, the evolution of anger is typically driven by two core factors: the driver's perceived pressure from the external environment and the discrepancy between their own expectations and reality. Choosing appropriate observations is crucial for properly describing the evolution of anger. This not only improves the model's applicability and generalization capabilities, but also provides more targeted emotion perception support for autonomous driving systems, thereby optimizing driving decisions.

[0087] The pressure from the external environment on the driver - headway:

[0088] The sense of oppression from the outside world on the driver is mainly reflected in the distance between the vehicle in front and the driver's own vehicle (headway). In traffic flow, when the distance between the vehicle in front and the driver's own vehicle is reduced, the driver's maneuverable space is limited, which can easily lead to increased tension or anxiety, and the continuous feeling of congestion may trigger anger. Therefore, headway can be used as a key variable to measure the impact of the driver's emotions on the external environment. In order to more accurately capture the driver's subjective feelings about the headway, considering the suddenness (mutation characteristics) of emotional changes, the present invention uses a 2-second time window to calculate the average headway, which is used as one of the input variables of the anger state recognition model. This time window can balance the short-term response to emotional changes with the continuous impact of environmental factors on the driver, ensuring that the model can effectively characterize the driver's sensitivity to external pressure.

[0089] The deviation between the driver's own expectations and reality - trajectory deviation:

[0090] In addition to external influences, the driver's anger is also closely related to the degree to which his or her own expectations are realized. When there is a large deviation between the driver's actual driving trajectory and his or her expected trajectory, it often causes frustration, and if this frustration continues to accumulate or cannot be alleviated, it may turn into anger. Therefore, trajectory deviation can be used as another important observation variable to measure the transition of the driver's anger. Taking into account that humans generally tend to minimize operating costs during driving (i.e., minimize unnecessary driving interventions as much as possible), the present invention adopts a short-time prediction method to estimate the driver's expected trajectory, and quantifies the driver's frustration by calculating the gap between the vehicle's actual trajectory and the expected trajectory. Specifically: a 2-second prediction window is set, that is, based on the current speed direction and initial velocity, the position where the driver expects to be in 2 seconds is predicted; the Euclidean distance between the actual trajectory point and the predicted trajectory point is calculated, and this is used as a measure of the degree to which the driver's expectations have not been met.

[0091] Example 1

[0092] In this embodiment, if Figure 1 As shown, a driver anger state warning method based on vehicle motion includes the following steps:

[0093] S1. Obtain the historical driving factors of anger emotion transition, including headway and trajectory deviation.

[0094] Space Headway is one of the important parameters for measuring the distance between the front and rear vehicles, and is closely related to the driver's following behavior, driving style and driving emotional changes. A shorter headway often reflects a higher sense of urgency and possible anger of the driver, while a longer headway may mean a more relaxed driving state. Therefore, headway can be used as a key observation to measure the driver's external pressure. Since the time resolution of the NGSIM dataset is 10 frames per second, this embodiment uses a time window of 2 seconds (i.e., 20 frames) to calculate the average headway. In this embodiment, the method for obtaining the headway includes: for each vehicle, in each time step, extracting the headway data of the vehicle closest to the front of the current vehicle; calculating the average of the headway data in the 2-second time window to obtain the headway:

[0095]

[0096] Among them, Avg_Space_Headway 2s represents the headway distance, N=20 represents the number of frames in a 2-second time window, t0 represents the time of the previous moment, and Space_Headway(t) represents the headway time of the current frame.

[0097] Predicted Position Deviation is used to measure the deviation between the vehicle's current trajectory and the driver's ideal trajectory. When the vehicle's actual trajectory deviates from the driver's expected route, it often causes frustration and anger in the driver. Therefore, this parameter can be used as an important indicator to measure the degree to which the driver's expectations are not met. In this embodiment, the method for obtaining trajectory deviation includes: selecting the current vehicle's position coordinates 2 seconds ago (X prev ,Y prev ) and driving speed v prev ; Based on the position coordinates (X prev ,Y prev ), calculate the current vehicle's driving direction θ 2 seconds ago:

[0098]

[0099] Based on the position coordinates (X prev ,Y prev ), driving speed v prev and driving direction θ, calculate the expected position (X pred ,Y pred ):

[0100] X pred =X prev +v prev ·2·cos(θ)

[0101] Y pred =Y prev +v prev ·2·sin(θ);

[0102] Get the actual position of the vehicle and based on the expected position (X pred ,Y pred )Calculate trajectory deviation:

[0103]

[0104] Among them, Deviation represents the trajectory deviation, X t Indicates the horizontal coordinate of the actual position of the vehicle, Y t The vertical coordinate representing the actual position of the vehicle.

[0105] S2. A Gaussian mixture hidden Markov model is constructed based on the Gaussian mixture model and the continuous hidden Markov model, and the Gaussian mixture hidden Markov model is trained based on the headway and trajectory deviation to obtain an anger state recognition model.

[0106] Currently, the HMM model in the existing technology mainly models a single observation sequence and is applicable to data samples containing only one variable. However, when it is necessary to simultaneously perform model training and classification prediction on data samples of two or more variables, the univariate HMM has certain limitations. The sample data involved in this embodiment includes two dimensions: the headway variable and the expected deviation variable. Therefore, the traditional single observation sequence HMM method is difficult to meet the modeling requirements. Therefore, it is necessary to introduce an HMM observation sequence framework suitable for multivariate scenarios to more accurately describe the joint time series characteristics of the headway and expected deviation.

[0107] When the training sequence set contains k training sequences, denoted as O, as follows:

[0108] O=[O (1) ,O (2) ,…,O (k) ]

[0109] Under the condition that each training sequence in the training sequence set is independent of each other, the goal of model parameter reestimation is to select a suitable model parameter λ so as to obtain the maximum value P(O|λ):

[0110]

[0111] Among them, K represents the number of iterations, P k represents the state probability of each iteration;

[0112] The model parameter reestimation formula is as follows:

[0113]

[0114] Among them, T k represents time k, represents the state probability, a ij represents the state transition probability matrix, b j represents the observation probability matrix, Indicates status, represents the observation probability, represents a certain observation state value, represents the observation probability, Represents a certain observation state.

[0115] In the parameters λ = (A, B, π) of the HMM model, B represents the probability distribution between the implicit state and the observation, which is usually expressed as a discrete multi-dimensional probability value array. However, in the application scenario of this embodiment, the goal is to use the headway and expected deviation variables to characterize the driver's emotional state, and these two variables are continuously distributed in the actual driving environment. This means that when identifying the observation data, the obtained headway and expected deviation values ​​will be any values ​​within a certain continuous range, and the traditional discrete probability array cannot accurately express the distribution characteristics of such continuous variables. Therefore, it is necessary to introduce a method that can describe the probability density of continuous observation variables in order to accurately characterize the time series changes of the headway and expected deviation.

[0116] After investigation, this embodiment adopts a continuous hidden Markov model (Continuous HMM) and uses multiple Gaussian distribution functions to fit the probability distribution B of the observation sequence. Theoretically, appropriate parameter selection can approximate the probability density distribution of any form, which is the core advantage of the Gaussian mixture hidden Markov model (GMM-HMM). The Gaussian mixture model (GMM) is a widely used statistical modeling method that can approximate complex probability distributions with a linear combination of multiple Gaussian distributions. GMM-HMM combines the time series modeling capability of HMM and the flexible probability density estimation capability of GMM, enabling it to more accurately describe the dynamic evolution of the driver's state. The observation probability distribution based on GMM can be calculated by the following formula, as shown below:

[0117]

[0118] In the GMM-HMM model, W represents the weight of different Gaussian distributions in the overall mixture model, that is, the mixing coefficient, which determines the contribution of each Gaussian distribution to the final probability density. At the same time, M represents the number of Gaussian distributions, P represents the dimension of the observed variables, and the p×p dimensional covariance matrix is ​​used to describe the shape and correlation of the Gaussian distribution. In addition, the transposed form of the covariance matrix is ​​also used to calculate the analytical expression of the probability density. Optimization is performed using the Maximum Likelihood Estimation (MLE) method to ensure that the constructed GMM-HMM can optimally fit the distribution of the observed data. In order to determine the final mixed Gaussian distribution, the following key parameters need to be estimated: mean vector, covariance matrix, and mixing coefficient. These parameters are usually optimized using the Maximum Likelihood Estimation (MLE) method to ensure that the constructed GMM-HMM can optimally fit the distribution of the observed data. Using the MLE method, the update formula for the optimization parameters can be derived, and the formula is as follows:

[0119]

[0120] Among them, λ represents the model parameter, P(x l |λ) represents the probability distribution under the model parameter λ, x l represents the lth data in the training data, N represents the number of elements in the training data, w j represents the jth Gaussian distribution, μ j represents the mean of the j-th Gaussian distribution, σ j represents the covariance of the j-th Gaussian distribution, Represents a multidimensional Gaussian distribution.

[0121] The Gaussian weights are estimated as:

[0122]

[0123] The mean is estimated to be:

[0124]

[0125] The variance is estimated as:

[0126]

[0127] Among them, γ represents the posterior probability, Z qli Indicates the sample value.

[0128] Therefore, the Gaussian mixture hidden Markov model in this embodiment includes: a hidden state and an observation state; the hidden state is used to represent the driver's anger level, including four discrete states:

[0129] S t ∈{S1,S2,S3,S4}

[0130] Among them, S t represents the anger state at time t; the observed state consists of the headway and trajectory deviation:

[0131] o t =[Avg_Space_Headway(t),Deviation(t)]

[0132] Among them, t represents the observed state of the vehicle at time t.

[0133] The training method includes: using the Baum-Welch algorithm to train the Gaussian mixture hidden Markov model; optimizing the model parameters by maximizing the log-likelihood of the observed data:

[0134]

[0135] in, Indicates the maximum value returned, O represents the sequence of vehicle observation states, and P(O|λ) represents the probability of each observation state occurring under the model parameter λ; the model parameters include: initial state probability π i , state transfer matrix A ij And the Gaussian mixture model parameters:

[0136] π i =P(S1=i)

[0137] A ij =P(s t+1 =j|S t =i)

[0138]

[0139] Among them, π i A represents the probability that the driver is in state i at time t=1. ij represents the probability of transitioning from state i to state j, P(o t |S t =j) represents the Gaussian mixture model parameters in state S t =j, Represents the probability density function of the multidimensional Gaussian distribution, M represents the number of Gaussian distributions, m represents the mth Gaussian distribution, w j,m represents the weight of Gaussian distribution m in state j, μ j,m represents the mean of the mth Gaussian distribution, Σ j,m represents the covariance of the mth Gaussian distribution, d represents the dimension, T represents the matrix transpose, μ represents the mean, and Σ represents the covariance matrix.

[0140] In this embodiment, the process of training the Gaussian mixture hidden Markov model using the Baum-Welch algorithm includes:

[0141] First, perform step E (calculate expectation):

[0142] α t (i)=P(O1,O2,…,O t ,q t =i|λ)

[0143] Among them, q t Indicates status;

[0144] Recursive calculation:

[0145]

[0146] Backward Probability:

[0147] β t (i) = P(Ot+1 ,O t+2 ,...,O T |q t =i,λ)

[0148] Recursive calculation:

[0149]

[0150] The posterior probability of the hidden state:

[0151]

[0152] State transition probability:

[0153]

[0154] Then perform M steps (parameter update):

[0155] Update the initial state probability:

[0156]

[0157] Update the state transition matrix:

[0158]

[0159] Update the B matrix:

[0160] Update the Gaussian mixture weights:

[0161]

[0162] Update the mean:

[0163]

[0164] Update the covariance matrix:

[0165]

[0166] Maximum likelihood of the last observation sequence:

[0167] P(O|λ)=Σ - {Q}P(O,Q|λ)

[0168]

[0169] S3. Use the anger state recognition model to monitor the driver and intervene in the driver's driving behavior when the anger level is detected to be higher than the preset threshold.

[0170] In this embodiment, based on the dual-dimensional observation of headway (external pressure) and trajectory deviation (expectation gap), the anger state recognition model is used to calculate the driver's anger level (S1-S4) in real time.

[0171] When the anger level is ≥ S3, graded interventions are triggered: Primary Intervention (S3): The central control screen displays a dynamic reminder "Please remain calm" and plays soothing music (heart rate synchronization detection verifies the emotional relief effect). Advanced Intervention (S4): Activate automatic driving takeover (3 seconds of progressive steering wheel return to center) and automatically open the windows 30% to introduce fresh air.

[0172] Furthermore, when the driver's anger level is ≥ S2, the navigation system automatically avoids sections requiring continuous lane changes (such as ramps in construction zones), prioritizing straight roads with ≥3 lanes. Combined with V2X data, the system broadcasts an "emotional driving warning" to surrounding vehicles (a red indicator projected onto the HUD of other vehicles) if the driver's anger persists. At anger level S4, the Electronic Stability Program (ESP) intervention threshold is reduced by 20% (preemptively mitigating the risk of skidding caused by aggressive steering).

[0173] Example 2

[0174] In this embodiment, a vehicle motion-based driver anger state warning system includes: a historical data acquisition module, a model construction module, and a monitoring module.

[0175] The historical data acquisition module is used to obtain the historical driving factors of anger emotion transition, which include: headway and trajectory deviation.

[0176] In the historical data acquisition module, the process of obtaining the headway distance includes: for each vehicle, in each time step, extracting the headway distance data of the vehicle closest to the current vehicle; calculating the average headway distance data within a 2-second time window to obtain the headway distance:

[0177]

[0178] Among them, Avg_Space_Headway 2s represents the headway distance, N=20 represents the number of frames in a 2-second time window, t0 represents the time of the previous moment, and Space_Headway(t) represents the headway time of the current frame.

[0179] In the historical data acquisition module, the process of obtaining trajectory deviation includes: selecting the position coordinates of the current vehicle 2 seconds ago (X prev ,Y prev ) and driving speed v prev ; Based on the position coordinates (X prev ,Y prev ), calculate the current vehicle's driving direction θ 2 seconds ago:

[0180]

[0181] Based on the position coordinates (X prev ,Y prev ), driving speed v prev and driving direction θ, calculate the expected position (X pred ,Y pred ):

[0182] X pred =X prev +v prev ·2·cos(θ)

[0183] Y pred =Y prev +v prev ·2·sin(θ);

[0184] Get the actual position of the vehicle and based on the expected position (X pred ,Y pred )Calculate trajectory deviation:

[0185]

[0186] Among them, Deviation represents the trajectory deviation, X t Indicates the horizontal coordinate of the actual position of the vehicle, Y t The vertical coordinate representing the actual position of the vehicle.

[0187] The model building module constructs a Gaussian mixture hidden Markov model based on the Gaussian mixture model and the continuous hidden Markov model, and trains the Gaussian mixture hidden Markov model based on the headway and trajectory deviation to obtain an anger state recognition model.

[0188] In the model building module, the Gaussian mixture hidden Markov model includes: hidden state and observation state; the hidden state is used to represent the driver's anger level and includes four discrete states:

[0189] S t ∈{S1,S2,S3,S4}

[0190] Among them, S t represents the anger state at time t; the observed state consists of the headway and trajectory deviation:

[0191] o t =[Avg_Space_Headway(t),Deviation(t)]

[0192] Among them, t represents the observed state of the vehicle at time t.

[0193] In the model building module, the training process includes: using the Baum-Welch algorithm to train the Gaussian mixture hidden Markov model; optimizing the model parameters by maximizing the log-likelihood of the observed data:

[0194]

[0195] Among them, λ represents the model parameter, Indicates the maximum value returned, O represents the sequence of vehicle observation states, and P(O|λ) represents the probability of each observation state occurring under the model parameter λ; the model parameters include: initial state probability π i , state transfer matrix A ij And the Gaussian mixture model parameters:

[0196] π i =P(S1=i)

[0197] A ij =P(s t+1 =j|S t =i)

[0198]

[0199] Among them, π i A represents the probability that the driver is in state i at time t=1. ij represents the probability of transitioning from state i to state j, P(o t |S t =j) represents the Gaussian mixture model parameters in state S t =j, Represents the probability density function of multidimensional Gaussian distribution, M represents the number of Gaussian distributions, m represents the mth Gaussian distribution, w j,m represents the weight of Gaussian distribution m in state j, μ j,m represents the mean of the mth Gaussian distribution, Σ j,m represents the covariance of the mth Gaussian distribution, d represents the dimension, T represents the matrix transpose, μ represents the mean, and Σ represents the covariance matrix.

[0200] The monitoring model uses the anger state recognition model to monitor the driver, and intervenes in the driver's driving behavior when the anger level is detected to be higher than the preset threshold.

[0201] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A driver anger state warning method based on vehicle motion, characterized in that: The following steps are involved: Obtaining historical driving factors of anger emotion transition, wherein the historical driving factors include: headway and trajectory deviation; A Gaussian mixture hidden Markov model is constructed based on a Gaussian mixture model and a continuous hidden Markov model, and the Gaussian mixture hidden Markov model is trained based on the headway and trajectory deviation to obtain an anger state recognition model; The anger state recognition model is used to monitor the driver, and when the anger level is detected to be higher than a preset threshold, the driver's driving behavior is intervened.

2. The method for warning driver anger based on vehicle motion according to claim 1, characterized in that: The method for obtaining the headway distance includes: For each vehicle, in each time step, extract the headway data of the vehicle closest to the current vehicle; The average of the headway data is calculated within a 2-second time window to obtain the headway distance: Among them, Avg_Space_Headway 2s represents the headway distance, N=20 represents the number of frames in a 2-second time window, t0 represents the time of the previous moment, and Space_Headway(t) represents the headway time of the current frame.

3. The method for warning driver anger based on vehicle motion according to claim 1, characterized in that: The method for obtaining the trajectory deviation includes: Select the current vehicle's position coordinates 2 seconds ago (X prev ,Y prev ) and driving speed v prev ; Based on the position coordinates (X prev ,Y prev ), calculate the current vehicle's driving direction θ 2 seconds ago: Based on the position coordinates (X prev ,Y prev ), the driving speed v prev and the driving direction θ, calculate the expected position (X pred ,Y pred ): X pred =X prev +v prev ·2·cos(θ) AND pred =And prev +p rev ·2·sin(θ); Get the actual position of the vehicle and based on the expected position (X pred ,Y pred ) Calculate the trajectory deviation: Among them, Deviation represents the trajectory deviation, X t Indicates the horizontal coordinate of the actual position of the vehicle, Y t The vertical coordinate representing the actual position of the vehicle.

4. The method for warning driver anger based on vehicle motion according to claim 1, characterized in that: The Gaussian mixture hidden Markov model includes: a hidden state and an observation state; The hidden state is used to represent the driver's anger level, including 4 discrete states: S t ∈{S1,S2,S3,S4} Among them, S t represents the anger state at time t; The observation state is composed of the headway and the trajectory deviation: o t =[Avg_Space_Headway(t),Deviation(t)] Among them, t represents the observed state of the vehicle at time t.

5. The method for warning driver anger based on vehicle motion according to claim 4, characterized in that: The training method includes: The Gaussian mixture hidden Markov model is trained using the Baum-Welch algorithm; Optimize model parameters by maximizing the log-likelihood of the observed data: Among them, λ represents the model parameter, Indicates the maximum value returned, O represents the sequence of vehicle observation states, and P(O|λ) represents the probability of each observation state occurring under the model parameter λ; The model parameters include: initial state probability π i , state transfer matrix A ij And the Gaussian mixture model parameters: π i =P(S1=i) A ij =P(s t+1 =j|S t =i) Among them, π i A represents the probability that the driver is in state i at time t=1. ij represents the probability of transitioning from state i to state j, P(o t |S t =j) represents the Gaussian mixture model parameters in state S t =j, Represents the probability density function of the multidimensional Gaussian distribution, M represents the number of Gaussian distributions, m represents the mth Gaussian distribution, w j,m represents the weight of Gaussian distribution m in state j, μ j,m represents the mean of the mth Gaussian distribution, Σ j,m represents the covariance of the mth Gaussian distribution, d represents the dimension, T represents the matrix transpose, μ represents the mean, and Σ represents the covariance matrix.

6. A driver anger warning system based on vehicle motion, the system applying the method according to any one of claims 1 to 5, characterized in that: include: Historical data acquisition module, model building module and monitoring module; The historical data acquisition module is used to obtain historical driving factors of anger emotion transition, and the historical driving factors include: headway and trajectory deviation; The model building module builds a Gaussian mixture hidden Markov model based on the Gaussian mixture model and the continuous hidden Markov model, and trains the Gaussian mixture hidden Markov model based on the headway and trajectory deviation to obtain an anger state recognition model; The monitoring model uses the anger state recognition model to monitor the driver, and when the anger level is detected to be higher than a preset threshold, the driver's driving behavior is intervened.

7. The driver anger state warning system based on vehicle motion according to claim 6, characterized in that: In the historical data acquisition module, the process of obtaining the headway distance includes: For each vehicle, in each time step, extract the headway data of the vehicle closest to the current vehicle; The average of the headway data is calculated within a 2-second time window to obtain the headway distance: Among them, Avg_Space_Headway 2s represents the headway distance, N=20 represents the number of frames in a 2-second time window, t0 represents the time of the previous moment, and Space_Headway(t) represents the headway time of the current frame.

8. The driver anger state warning system based on vehicle motion according to claim 6, characterized in that: In the historical data acquisition module, the process of obtaining the trajectory deviation includes: Select the current vehicle's position coordinates 2 seconds ago (X prev ,Y prev ) and driving speed v prev ; Based on the position coordinates (X prev ,Y prev ), calculate the current vehicle's driving direction θ 2 seconds ago: Based on the position coordinates (X prev ,Y prev ), the driving speed v prev and the driving direction θ, calculate the expected position (X pred ,Y pred ): X pred =X prev +v prev ·2·cos(θ) AND pred =And prev +v prev ·2·sin(θ); Get the actual position of the vehicle and based on the expected position (X pred ,Y pred ) Calculate the trajectory deviation: Among them, Deviation represents the trajectory deviation, X t Indicates the horizontal coordinate of the actual position of the vehicle, Y t The vertical coordinate representing the actual position of the vehicle.

9. The driver anger state warning system based on vehicle motion according to claim 6, characterized in that: In the model building module, the Gaussian mixture hidden Markov model includes: a hidden state and an observation state; The hidden state is used to represent the driver's anger level, including 4 discrete states: S t ∈{S1,S2,S3,S4} Among them, S t represents the anger state at time t; The observation state is composed of the headway and the trajectory deviation: o t =[Avg_Space_Headway(t),Deviation(t)] Among them, t represents the observed state of the vehicle at time t.

10. The driver anger state warning system based on vehicle motion according to claim 9, characterized in that: In the model building module, the training process includes: The Gaussian mixture hidden Markov model is trained using the Baum-Welch algorithm; Optimize model parameters by maximizing the log-likelihood of the observed data: Among them, λ represents the model parameter, Indicates the maximum value returned, O represents the sequence of vehicle observation states, and P(O|λ) represents the probability of each observation state occurring under the model parameter λ; The model parameters include: initial state probability π i , state transfer matrix A ij And the Gaussian mixture model parameters: π i =P(S1=i) A ij =P(s t+1 =j|S t =i) Among them, π i A represents the probability that the driver is in state i at time t=1. ij represents the probability of transitioning from state i to state j, P(o t |S t =j) represents the Gaussian mixture model parameters in state S t =j, Represents the probability density function of multidimensional Gaussian distribution, M represents the number of Gaussian distributions, m represents the mth Gaussian distribution, w j,m represents the weight of Gaussian distribution m in state j, μ j,m represents the mean of the mth Gaussian distribution, Σ j,m represents the covariance of the mth Gaussian distribution, d represents the dimension, T represents the matrix transpose, μ represents the mean, and Σ represents the covariance matrix.

Images